Meta-analysis of 2040 sickle cell anemia patients: BCL11A and HBS1L-MYB are the major modifiers of HbF in African Americans.

Fetal hemoglobin (HbF) protects against many but not all of the hematologic and clinical complications of sickle cell anemia.1,2 This protection is dependent on the ability of HbF to hinder deoxyHbS polymerization. HbF level is variable and highly heritable. Previous genetic association studies found single nucleotide polymorphisms (SNPs) in regions of BCL11A (chromosome 2p), in the HBS1L-MYB intergenic polymorphism (HMIP; chromosome 6q), and linked to HBB (chromosome 11p) that were associated with HbF (reviewed in Akinsheye et al1). Our aim was to perform a meta-analysis of genome-wide association studies (GWAS) to find genetic loci with modest effect sizes that were associated with HbF when a larger sample size was examined. Common SNPs (585, 563 total) from 7 cohorts totaling 2040 patients were meta-analyzed using the software Meta Analysis Helper (METAL)3 with inverse variance method, where effect estimates are weighted in proportion to their precisions.4 The 7 cohorts included in the meta-analysis are: Cooperative Study of Sickle Cell Disease (CSSCD: n 841), Multicenter Study of Hydroxyurea (MSH: n 178), Pulmonary Hypertension and the Hypoxic Response in Sickle Cell Disease (PUSH) study (n 73), Comprehensive Sickle Cell Centers Collaborative Data (C-data) project (n 127), Treatment of Pulmonary Hypertension and Sickle Cell Disease with Sildenafil Treatment (Walk-PHaSST) trial (n 181), Duke University Outcome Modifying Genes study (n 152), and Silent Infarct Transfusion (SIT) trial (n 488). In each of these studies, patients with the HbS only phenotype, aged 5 years or more when HbF was measured were included. None of the patients were treated with hydroxyurea when HbF was measured. The association between HbF and the genotype for each SNP was tested in a multiple linear regression analysis, adjusting for sex and the top 10 principal components, where appropriate. The additive genetic model, which codes the SNP genotype as the number of minor alleles (0,1,2), was assumed. To obtain a normal distribution of the phenotype for analysis the cubic root of HbF was used as the response variable in the regression analysis, as previously established.5 Table 1 summarizes the results of the meta-analysis for the SNPs that reached genome-wide significance (P 5 10 8). Consistent with previous findings, the most significant SNPs were in BCL11A (lowest P value of 5.36 10 58 for rs766432). In addition, 3 SNPs in HMIP reached genome-wide significance. This gene is situated in the interval between the gene HBS1L (a G-protein/elongation factor) and the MYB an erythroid transcription factor, on chromosome 6q23.3, and was previously shown to be associated with HbF in a microarray study.6 SNPs in OR51B5, which were associated with HbF in a previous GWAS,5 did not reach genome-wide significance. These SNPs almost reached genome-wide significance in the CSSCD, but were not significant in other cohorts at .05 (supplemental Table 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online letter). In the CSSCD cohort, the most significant SNPs in BCL11A (rs766432) and the HMIP region (rs9494145) explained 11.1% and 3.2% of the phenotypic variability in HbF, respectively, and together explain only 14.7% of the variability. Twelve additional SNPs reached statistical significance of 10 5 or less, although none reached genome-wide significance (supplemental Table 1). HbF is regulated as a complex trait and the “missing heritability” not detected by GWAS has various explanations. These include unaccounted gene interactions, large undetected insertions and deletions, epigenetic factors, regulation by small RNAs, and multiple rare variants with small effects.7 For example, SNP rs2033467 in chromosome 5 (supplemental Table 1) is in a novel region but it did not reach genome-wide significance, although it was statistically significant.